The Chrome DevTools team shipped chrome-devtools-mcp — an official Model Context Protocol server that lets coding agents (Gemini, Claude, Cursor, Copilot) control and inspect a live Chrome browser. It gives agents the full DevTools toolkit for automation, debugging, and performance analysis.
## What an agent can do with it
Record performance traces and extract actionable insights. Analyze network requests. Take screenshots. Read browser console messages — with source-mapped stack traces, so the agent sees the original source location, not minified gibberish. Automate actions via Puppeteer, with automatic waiting for results. In short: the agent gets eyes and hands inside the browser, not just the DOM.
## The autoConnect feature
By default the server spins up a fresh Chrome instance with a dedicated profile. But autoConnect (stable since Chrome 146) lets it attach to your existing browser session — so an agent can debug the exact page you’re looking at, sharing state between manual and agent-driven testing. Distributed via NPM.
## Why it matters
Frontend debugging has been the blind spot for coding agents — they could write the code but couldn’t see what the browser actually rendered or why it broke. An official Chrome-team MCP closes that gap. Combined with the recent wave of agent-browser tools, the browser is quickly becoming a first-class agent surface.

Leave a comment